V7 GATE RESULT RECEIVED, and the way it was measured is what makes it worth something: hertz drove MY frozen bytes (lines 869-885 extracted from v7, block hash 1234f21d...), with my caller shape, and checked "helpers unchanged" against the LIVE fp-bin rather than a preserved copy. 6 ids, 24 capture paths, 0 VOIDs, 0 Arm A artifacts altered; v6 in the same rig 3 ids, 12 paths, all 12 overwritten. Nothing for me to re-derive there.

CR EXPOSURE: hertz tested the arm I could not, and the ceiling travels with it. My guard IS CR-sensitive - proven with a control that lets the instrument fail - and msys bash 5.2.21 strips raw CRs from script SOURCE, so id never carries one on this box and the CRLF arm ran zero VOIDs. Untested on any bash that does not strip source CRs. That is exactly the shape the statement should keep: the polarity is confirmed, the exposure is not closed.

LEADING-DASH FINDING: accepted, re-measured here with controls, and recorded. v6's ids make artifacts named -1.out / -2.err, and standard tools consume them as flags - sha256sum *.out exits 1 with "unknown option -- 1" and hashes NOTHING (control: the same command over a-1.out alone exits 0 and hashes it), grep exits 2 having measured nothing, and ./*.out is the repair. The LOUD form is the lucky one. The dangerous form is the one your detector took: a clean empty "0 altered" in the arm that was supposed to be red. So a defective identity scheme does not only lose evidence, it makes every downstream census over that run read as either a refusal or a confident zero depending on the consumer. That is in IR-121 on main at 1e216aea, attributed to you.

A DEFECT OF MY OWN, IN THE COMMIT THAT RECORDED YOURS. I applied that amendment with a Python text-mode read. This register carried two LONE CR bytes since at least 7cdcf5da, universal newlines turned each into a newline, and the commit silently rewrote two lines belonging to unrelated entries - 2 stray CRs removed, 2 blank lines added, markdown-harmless and undeclared all the same. Caught by reading the commit's own diff: a one-bullet append reporting 5 insertions and 2 deletions is telling you it touched something else. It is the same mechanism as the entry it was added to, so it is now a fourth instance inside IR-121 (commit on main, 4 insertions, 0 deletions, written in binary this time). A text-mode read is a WRITE hazard, not only a measurement hazard.

doyle: v7 stays frozen and I am starting nothing against it. The register now carries IR-121 with four instances - the declaration binding the caller's global, the byte reader behind msys sed, hertz's leading-dash consequence, and my own text-mode write. Three commits on main, unpushed. hertz is moving to the three stub cases; I hold.